Towards a reliable prediction of conversion from Mild Cognitive Impairment to Alzheimer s Disease: stepwise learning using time windows
|
|
- Beverly Baker
- 5 years ago
- Views:
Transcription
1 Towards a reliable prediction of conversion from Mild Cognitive Impairment to Alzheimer s Disease: stepwise learning using time windows Telma Pereira * * INESC-ID and Instituto Superior Técnico, Universidade de Lisboa, Portugal telma.pereira@tecnico.ulisboa.pt Francisco L. Ferreira * francisco.lourenco.ferreira@tecnic o.ulisboa.pt Manuela Guerreiro Laboratório de Neurociências, IMM and Faculdade de Medicina, Universidade de Lisboa, Portugal mmgguerreiro@gmail.com Alexandre de Mendonça mendonca@medicina.ulisboa.pt Sara C. Madeira + + LASIGE, Faculdade de Ciências, Universidade de Lisboa, Portugal sacmadeira@ciencias.ulisboa.pt for the Alzheimer s Disease Neuroimaging Initiative 1 ABSTRACT Predicting progression from a stage of Mild Cognitive Impairment to Alzheimer s disease is a major pursuit in current dementia research. As a result, many prognostic models have emerged with the goal of supporting clinical decisions. Despite the efforts, the clinical application of such models has been hampered by: 1) the lack of a reliable assessment of the uncertainty of each prediction, and 2) not knowing the time to conversion. It is paramount for clinicians to know how much they can rely on the prediction made for a given patient (conversion or no conversion), and the time windows in case of conversion, in order to timely adjust the treatments. We propose a supervised learning approach using Conformal Prediction and a stepwise learning approach, where the learning model first predicts whether a patient converts to dementia, or remains stable, and then predicts the more likely progression window (short-term or long-term conversion). We used data from ADNI to test the approach and predict conversion within time windows of up to 2 years (short-term converter) and 2 to 4 years (long-term converter). The exploratory results are promising but compromised by the small number of examples for the long-term converting patients, available for training. Keywords Conformal Predictors, Confidence Estimation, Mild Cognitive Impairment, Alzheimer s Disease, Prognostic Prediction 1 1 Data used in preparation of this article were obtained from the Alzheimer s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: f 1. INTRODUCTION Alzheimer's disease (AD) is a neurodegenerative disease with devastating effect on patients and their families, and a huge socioeconomic impact in modern societies. Nowadays, more than 30 million people suffer from AD worldwide and its prevalence is expected to triple by 2050 [18]. It is therefore paramount to understand AD and its progression, not only to guide clinical decisions and manage patients and families expectations, but also to develop new effective treatments. Mild Cognitive Impairment (MCI) is considered as a transitive stage between healthy aging and dementia [18], where patients have cognitive complaints not interfering with daily live activities [18]. These patients are more likely to develop AD [11]. In this context, studying progression from MCI to dementia is a major challenge in current medical research [1, 21]. Whilst there is no treatment to revert AD s symptoms, an early prognostic of dementia is nevertheless useful to help clinicians taking decisions about their patients possible treatment and to timely adjust medical appointments. By following different approaches and using different types of data (biological markers and/or neuropsychological data), researchers have sought for robust prognostic models, to guide clinical decisions by means of a medical decision support system to be used in clinical settings. This clinical decision support system would predict the most likely prognostic for a new MCI patient based on the past history of a cohort of patients with known diagnostics. Despite the advances made in prognostic prediction for MCI patients, where machine learning techniques achieved promising results [13], some issues have hampered its practical use in clinical settings: 1) Lack of trustworthy prognostic prediction. Most prognostic models only produce bare predictions, without providing any assessment of the uncertainty on each prediction. This is a major disadvantage as for clinicians it is paramount to know
2 how much can they trust the prognostic predicted for a given patient [16, 19]. In this context, confidence measures can provide insight on the likelihood of each prediction. 2) Unknown time to conversion. The early prognostic of AD is very important to treat patients as well as possible and manage their expectation regarding disease progression. Furthermore, knowing how fast will the progression be in case it occurs is of great value. With such information, clinicians could timely adjust treatments and medical appointments to the need of that specific patient. In this context, we should model conversion to dementia within specific time intervals. Besides being more informative from a clinical point of view, the few studies [2, 7, 17] that addressed this question showed that prognostic models learned with patients having similar time to conversion are more reliable than those models learned from heterogeneous groups of patients (regarding their conversion time). In this work, we propose a supervised learning approach aiming to tackle these issues. Following a stepwise learning process, this approach not only predicts conversion to dementia, but outputs the more likely window of occurrence (short-term or long-term conversion). To test this approach, we used Conformal Prediction (CP) [20, 23], a machine learning technique that produces confidence measures. Conformal Predictors (CPs) are built on top of traditional machine learning algorithms (denoted as underlying algorithms), predicting the class that makes the new example (patient) more conform to the training set within certain levels of confidence. A confidence level of 0.9, for instance, means that the conformal predictors commit to a maximum of 10% of errors. CPs have been applied successfully in disease-related problems such as the early detection of ovarian cancer [5], diagnosis of acute abdominal pain [15] or stroke risk assessment [10]. Neuropsychological data have proved their relevance on predicting converting MCI patients, being considered as accurate as the more complex models involving data integration [12]. Despite the value of biological biomarkers, they retain a supportive role to the neuropsychological assessment. On the one hand, measurable cognitive impairment is still a hallmark for the diagnosis of dementia and mild cognitive impairment [6, 24]. On the other hand, neuropsychological tests (NPTs) are less expensive, non-invasive, and easily applied in clinical practice. Machine learning approaches are gaining a lot of relevance in dementia research [2, 13], but mainly focusing on brain imaging data (Magnetic Resonance Imaging (MRI) or Positron Emission Tomography (PET)). Works involving only NPTs tend to use traditional statistical analysis, which may not be suitable to successfully capture all its predictive power [1]. In this work, we study the feasibility of the proposed approach to the prognostic problem of MCI-to-AD conversion within certain time windows, using the conformal prediction framework and neuropsychological data. To our knowledge, this was not explored to date. The proposed supervised learning approach may, however, be used with other methods and/or data types. 2. DATASET AND METHODS 2.1 ADNI data The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. Its goals is to find relevant biomarkers in all stages of AD to guide future clinical trials for new possible treatments. Currently, it includes 1400 MCI and AD patients, as well as normal subjects, being followed. ADNI includes several biomarkers of Alzheimer s disease such as demographic data, neuropsychological tests, Cerebrospinal fluid (CSF), structural Magnetic Resonance Imaging, functional-mri (fmri), Positron Emission Tomography and other biological data. This data is collected from every ADNI participant at the baseline assessment, as well as on their annual follow-up consultations. In this work, we used NPTs data from ADNI-2 patients. A list of the 93 features used (original or created from the data) is available at: This data also includes demographic data such as age, gender, education level. NPTs include, but are not limited to, the Mini Mental State Examination (MMSE) and the Alzheimer s Disease Assessment Scale cognitive subscale (ADAS-Cog). For a detailed description of ADNI s data, we refer to The total number of patients was filtered to include only MCI patients at the baseline assessment. Reverters patients that converted from MCI to normal cognition, or even AD to MCI, were excluded from the pool. This is usually the method employed by most studies, as reversion is clearly unexpected and can be related to errors in diagnosis or the presence of diseases other than AD [8], which affects the diagnosis. Also, patients were excluded if they didn t have any follow-up assessments. We followed the strategy to create learning examples using time windows described in [17]. For a given time-window, we considered patients that converted to dementia within a predefined interval, i.e. which had the diagnosis of AD in one of the yearly assessments up until the limit of the window. Those are labeled cmci (converter MCI). On the other hand, patients that didn t convert to AD during that period and presented a diagnosis of MCI at the limit of the window or afterwards, are included in the learning set labelled as smci (stable MCI). In this work, and taking into account the nature of the ADNI data, we chose a follow-up time of 4 years. From a total of 265 MCI patients, 143 (54%) patients remained stable and 122 (46%) converted to dementia within the follow-up period. Regarding the converting patients, 89 converted to AD within the first 2 years (short-term conversion, s.t. cmci), and 33 converted between 2 and 4 years (long-term conversion, l.t. cmci). 2.2 Conformal Prediction We introduce the idea behind the conformal prediction framework. For a more formal description we refer to [23]. Let us assume that we are given a training set {(! ", $ " ),...,(! "$, & "$)}, where! " $ is a vector of attributes and! " $ is the class label (assuming a binary classification problem). Given a new test example (" ) we aim to predict its class. Intuitively, we assign each class! " $ to! ", at a time, and then evaluate how nonconform the example (! ", $ " ) is in comparison with the training data. The most likely class label conforms better with the training set. A non-conformity measure, to assess the non-conformity of the test example, must be extracted from the underlying classifier (any classifier may be used). To evaluate how different! " is from the training set, we compare its non-conformity score with those of the remaining training examples! ", % = 1,, ) 1, using the p-value function (distinct from the p-value from statistics):! " = '(),,: -. /- 0 (1)
3 where! " is the non-conformity score of! ", assuming it is assigned to the class label! ". If the p-value is small, then the test example (! ", $ " ) is non-conforming since few examples (! ", $ " ) had a higher non-conformity score when compared with! ". On the other hand, if the p-value is large,! " is very conforming since most of the examples (", % ) had a higher non-conformity score when compared with! ". Once p-values are computed, CP can be used in one of the following ways: 1) Using prediction regions. For a given significance level (e), CPs output a prediction region,! " : set of all classes with! " > &, contrarily to the single predictions given by standard classifiers. These prediction regions have a guaranteed error rate. This means that the frequency of errors (fraction of true values outside! " ) does not exceed!, at a confidence level 1. The error rate is guaranteed under the randomness assumption, which states that the examples are independently drawn from the same distribution (this property is called validity) [23]. Prediction regions may therefore comprise more than one class (uncertain prediction), no class (empty prediction) or a single class (certain prediction). Multiple predictions are not errors but a reflection of the fact that the classifier is not being confident enough to predict a certain class. The smaller the prediction region the more efficient the conformal predictor is [23]. 2) Using forced predictions. If one would rather have single predictions than prediction regions, CPs predict the class with the highest p-value (forced prediction), alongside with its credibility (the largest p-value) and confidence (complement to 1 of the second highest p- value). Confidence reveals how likely the predicted classification is compared with the other classes. Credibility reveals how suitable the CP is for classifying the given example. Low credibility means that either the training set is non-random or the test example is not representative of the training set, and thus, the predicted class is non-conforming to the training data. Given that the data was generated independently from the same distribution (randomness assumption [23]), the probability that the credibility is less than some threshold e is less than e [22]. The higher the values of both confidence and credibility the more reliable is the prediction Transductive and Inductive CP Conformal prediction may be used in the transductive or in the inductive setting. When transductive framework is used, the training set is enriched with the test example, and the underlying classifier is updated. Non-conformity scores are then computed for all the training examples. This process is repeated for all class labels!. A new prediction is therefore based on all the training examples. For large datasets, this is computationally very demanding. This led to the emergence of inductive learning [14, 22]. When inductive is used, the training set {(! ", $ " ),...,(! "$, & "$)} is divided into the proper training set {(! ", $ " ),...,(! ", $ ")} and the calibration set {(! "$, & "$ ),...,(! "$, & "$)}, where! < 1. The proper training set is used to derive the prediction rule, by training the underlying classifier. This prediction rule is then used to classify the examples of the calibration set and the test example. Nonconformity scores are only computed with the examples of the calibration set Mondrian CP Mondrian conformal prediction is a variant of CP that deals with imbalanced datasets [23]. When the number of examples of a given class is significantly larger than those of the other class, most errors are putatively from the minority class, limiting the applicability of these predictions. Mondrian conformal prediction applies CPs separately to each label class. The p-value is thus computed by comparing the non-conformity score of the test example against only training examples of the same class as the current hypothesis! " :! " = '(),, -.(- / / '(),, -. (- / (2) 2.3 Stepwise learning with time windows The supervised learning approach proposed in this work is described in Figure 1. It consists on a two-step supervised learning approach which starts by predicting conversion from MCI to AD, within a given level of confidence, and then complements it with the prediction of the most likely time window of conversion (short-term or long-term conversion). More specifically, in the first step, Model 1 predicts whether a given MCI patient will convert to dementia (cmci) or remains MCI during the follow-up period (smci). A measure reflecting the confidence on the predicted class is outputted. If this value is below a certain (predefined) threshold, we consider that prediction as unpredictable (No prediction). If the prediction is trustworthy (above the confidence level) the prognostic is made. A trustworthy prediction of conversion (cmci) is fed into Model 2, in the second step of the approach. This model predicts whether the patient will have a short-term or long-term conversion. Once again, low confident predictions are considered as unpredictable. Figure 1. Workflow of the proposed stepwise supervised learning approach. Model 1 predicts whether a MCI patient is going to convert (cmci: converter MCI), remains MCI (smci: stable MCI), or if no prognostic is presented because the prediction is below the chosen confidence level. Model 2 predicts the more likely window of conversion (short-term or long-term) for converting patients, or no prediction if it is not trustworthy Classification setup The proposed classification approach is divided into two phases: i) training and tuning the parameters to Model 1 (prognostic prediction, Figure 1) and Model 2 (window of conversion, Figure 1) and ii) applying the stepwise approach described before: for a
4 given test set, first classify patients as smci or cmci and then in short-term (cmci at 0-2 years) or long-term (cmci at 2-4 years) conversion. The dataset was randomly split (keeping class proportions) in training set (80%) and test set (20%). This process was repeated 5 times with fold randomization. During the training and tuning parameters phase, for each training set, a further 5-fold cross-validation (CV) procedure was repeated 5 times with fold randomization. This aimed at dividing the examples into training and validation sets, in order to use the validation examples to determine the parameters that optimize a predefined evaluation metric. We optimized the values of F-measure and outputted also the values of sensitivity and specificity. We used Naïve Bayes and tested three parameters: 1) use Gaussian distribution or 2) kernel estimator for numerical attributes, and 3) whether to use supervised discretization to convert numerical into nominal attributes. Since the use of preprocessing techniques to deal with a large number of (possibly irrelevant) features or imbalanced classes may have a significant impact on both classification performance and model simplification and interpretability, the worth of using/not using feature selection (FS) and/or using/not using SMOTE to deal with class imbalance was tested. Four methods from the filter family of FS were tested. A filter feature selection algorithm evaluates the value of a features subset without taking into account the learning algorithm that is applied afterwards. These methods take different measures in search for the most predictive features, such as Weka s CfsSubsetEval is an implementation of Hall s work [9], that measures the predictive power of each feature while minimizing the redundancy between them. CorrelationAttributeEval measures the Person s correlation between features and the class and SymmetricalUncertAttributeEval and InfoGain measure the worth of a particular attribute by its information gain and symmetrical uncertainty with respect to the class, respectively. The latter three methods rank the features by their individual evaluations and thus the number of selected features to keep must be defined. Subset sizes of 10 to 40 features, with incremental steps of 5 features, were tested. These FS methods were implemented in Weka. Furthermore, class imbalance was tackled with the Synthetic Minority Over-sampling Technique (SMOTE) [3]. SMOTE is an oversampling technique that generates synthetic samples from the minority class by choosing a set of similar instances and perturbing the attributes by a random amount. SMOTE percentage ranges from 0% to the equilibrium of the class proportions, in 3 steps. In order to ensure the validity of the results, FS and SMOTE were only applied to the training data within each crossvalidation fold. The worth of using FS and/or SMOTE was assessed by the Wilcoxon Signed Rank Test [4] on the averaged F-Measure across the 5 5-fold CV using IBM SPSS Statistics 24 (released version ). After the tuning phase, we trained Model 1 and Model 2 with the training data (80% of the original dataset) and the optimized parameters and then, we tested the proposed stepwise approach with the test set (20% of the original dataset). The classification approach was implemented in Java using WEKA's functionalities (version 3.8.0) Conformal prediction settings Given that the dataset under study does not have high dimensionality, we used the Transductive Conformal Prediction framework. In addition, we used Mondrian CP framework to tackle the class imbalance of the dataset under study. We used Naïve Bayes as underlying classifier of the CP approach since, in previous work [17], it outperformed other commonly used classifiers (such as SVMs, Decision Trees and Random Forests) in the MCI-to-dementia conversion problem. The following non-conformity measure was used: - log % & ' = ) * ', (3) where! is the posterior probability estimated by Naïve Bayes. We used the Forced Prediction approach to train and tune the parameters to Model 1 and Model 2 (first phase of the proposed approach, Section 2.3.1). The confusion matrix computed to finetune the parameters was built with no confidence level as it used all the (forced) predictions made. In the second phase of the proposed framework (applying the stepwise approach), both approaches of Conformal Prediction using Prediction Regions (PR approach) or Forced Predictions (FP approach) (section 2.2) were used. In the former, certain predictions were considered as trustworthy while uncertain and empty predictions were seen as unreliable (no class predicted). Three significance levels were tested! = {0.15, 0.20, 0.30} corresponding to the following confidence levels: cfd = {0.70, 0.75, 0.85}. In the latter, forced predictions above the predefined confidence threshold were considered as trustworthy while the remaining were disregarded. Three confidence (cfd) thresholds were tested (cfd = {0.70, 0.75, 0.80}). Two other approaches, using Prediction Regions, were tried to compute the optimization metric in the parameters tuning phase (first phase of the proposed approach, Section 2.3.1). Specifically, F-measure was computed: i) using only certain predictions (disregarding the uncertain and empty predictions) and ii) using the certain predictions and considering the uncertain and empty predictions as misclassified examples. However, since these approaches did not enhanced the results and due to space constraints, we do not report these results. 3. RESULTS AND DISCUSSION The data used in this work is described in Section 2.1 and summarized in Table 1. Table 1. Data description. smci cmci Model (54%) 122 (46%) Model 2 - s.t. cmci l.t. cmci 89 (73%) 33 (27%) From the empirical experiments performed using different FS methods and size of the features subset, we chose the InfoGain with 25 features. Small differences were found between the averaged F-Measure with distinct FS methods. Although good results were achieved for larger subsets of features, we decided to pursued the analysis with 25 features, as it represents a good trade-off between a good classification performance and a minor number of features used, which should be as small as possible regarding the reduced sample size.
5 When evaluating the worth of using FS with Model 1, we concluded that there was no statistical difference (p-value<0.291) between the results obtained with or without feature selection. Still, we decided to proceed with feature selection for the sake of model interpretability. We note that SMOTE was not applied to Model 1 since its respective class imbalance is negligible. In what concerns Model 2, using SMOTE enhanced the results (pvalue=0.00), either using or not using FS. Moreover, learning the model with a reduced set of features also improved the results (pvalue<0.04). As such, we pursued the analysis with FS for both Model 1 and Model 2 and with SMOTE for Model 1. As aforementioned, InfoGain ranks features according to their information gain in respect to the class. As such, and while not being the main goal of this work, we briefly evaluated the highest and lowest ranked features for Model 1 and Model 2. Table 2 reports the highest and the lowest 10 features for each model. Extended tables are provided in the following link: ADAS-Cog total scores and memory sub-scores (word recall and delayed recall) along with the total score of the Functional Assessment Questionnaire are amongst the highest ranked features of Model 1. Different measures of the Clinical Dementia Rating are also present, including the memory subscore but also measures of daily live activities, judgment, problem solving and orientation. Regarding the highest ranked features of Model 2, we can see subscores from the Neuropsychiatric Inventory that measures several behavioral patterns such as the patient s appetite, depression, irritability, apathy and sleep. These were totally absent from the highest ranked features of Model 1. Although there are common highest ranked features for both Models (such as sub-scores of ADAS-Cog and the Clinical Dementia Ranting), there are some measures, such as total scores of ADAS-Cog, part B of the trail making test and the total score of the Functional Assessment Questionnaire that appears as a top feature for Model 1 but as one of the bottom features for Model 2. These results suggest that while some measures of delayed recall are extremely important as Alzheimer s disease biomarkers, there are marked differences in the most predictive tests and scores to classify converting patients and differentiate early from late converters. Finally, some features ranked low on both Models, such as the patient s age, education level, number of years of symptoms prior to baseline and age at retirement. Table 3 reports the results obtained with the optimized parameters tuned within a randomized 5-fold cross validation scheme, for each train set (5 80% data) and for Model 1 and Model 2. As aforementioned, Model 1 predicts whether a MCI patient converts to AD (positive class, cmci) or remains stable during the followup period of 4 years (negative class, smci). The model successfully distinguished between converting and non-converting patients, as values above 0.76 were achieved in all evaluation metrics. Non-converting patients were easier to identify than those who converted, pointed by the higher values of specificity. Table 3. Results obtained with the optimized parameters finetuned within a randomized 5-fold cross validation scheme, for each train fold (5 80% data) and for each model 1 (prognostic prediction) and 2 (time to conversion). F-Measure Sensitivity Specificity Model ± ± ±0.014 Model ± ± ±0.029 In the second step of the stepwise approach, Model 2 aims to predict the most likely window of conversion, between a shortterm (up to 2 years) and long-term (2-4 years) conversion. This classification task is more challenging than the prognostic prediction learned by Model 1. In fact, the classification models are based on the performance accomplished in the neuropsychological assessment made at the baseline (first patient s assessment). A converter MCI has certainly a more accentuated deficit in the NPTs when compared with a stable MCI. Between converting patients, differences in the performance on the NPTs assessment should also be present, since some patients are closer to become demented than others, although at a smaller level, which hampers the learning task. The results corroborate this idea (Model 2, Table 3). The negative class corresponds to a short-term conversion while the positive class represents the long-term conversion. Despite the difficulty of this classification task, and the small and imbalanced set of examples used in the learning process, the performance was above the random level, with F-Measure of 64% and equilibrated values of sensitivity and specificity (63% and 62%, respectively). Tables 4 and 5 report the results obtained with the stepwise approach ran with the test set (5 20% data), using the models trained with the optimized parameters, and following the Forced Prediction and Prediction Regions approaches (Section 2.3.2), respectively. We note that, since Models 1 and 2 learn distinct classes (Model 1: smci vs cmci and Model 2: short-term cmci vs long-term cmci) we cannot present an overall evaluation metric to assess their performance (as we did in Table 3). Instead, we evaluate the number of cases that are correctly (and incorrectly) classified for each class and at each step of the stepwise approach, at different confidence thresholds. We note that the class distribution in the test set is imbalanced (it has in average 24 smci, 18 short-term cmci and 6 long-term cmci) thus hampering the results, mainly during the second step. Model 1 has a high predictive power on classifying MCI patients as being converting or non-converting, following both approaches, with about 84% of non-converting cases correctly classified, even with the highest confidence threshold (cfd=0.80) for the FP approach and 75% for the PR approach. About 79 to 83% (cfd = 0.70 to 0.80) of the converting cases were also correctly identified, although the number of cases with short-term conversion that were correctly classified were much higher (above 83%) than those who converted between 2 and 4 years (around 60%), for the FP approach. This may be, however, an effect of the imbalanced dataset, and not of the incapability of this class to be learned. As the confidence threshold increases, the number of unreliable predictions (number of non-classifiable cases) also increases from 0 to 6%. Still, even for high values of confidence, Model 1 achieved good classification performances and a small number of non-classifiable cases. Similar results were obtained with the PR approach. In this scenario, there is however a higher amount of untrustworthy predictions (up to 32%) but also a smaller number of wrong predictions (maximum of 12%). A final prognostic prediction is given for non-converting patients (Model 1). Those classified as converters (about 16 short-term cmci and 3 long-term cmci) are then fed to Model 2, to complement their prognostic with a likely window of progression. About 64% of the short-term converting patients are correctly classified with a confidence level of 0.70 for the FP approach. On the other side, this model was not reliable on identifying longterm converters since it correctly classified at maximum average
6 of 1.4 cases out of 3.8. Once again, the small number of long-term cmci does not allow us to evaluate whether these results are due to the challenging learning task or a consequence of the class imbalance. Still, it is clear that this classification task is harder than that learned with Model 1. The predictions are in general less confident as the number of non-classifiable cases goes from 6 to 32%, with the growth of the confidence threshold. Comparing with the PR approach (Table 5) a minor number of cases of s.t. and l.t. conversions were predicted, mainly for higher confidence levels, but also less errors were made (11% for confidence level of 0.85). This is due to the larger amount of non-classifiable patients obtained with this approach. This leads to the idea that the PR approach is more stringent on the demanded trustiness to make a prediction than the FP approach. In fact, the FP approach outputs a larger number of correctly classified cases but also a higher number of misclassified cases, as only a reduced number of cases are considered as non-classifiable. 4. CONCLUSIONS AND FUTURE WORK This paper presents a two-step supervised learning approach which starts by predicting conversion from MCI to AD, within a given level of confidence, and then complements it with the prediction of the most likely time window of conversion (shortterm or long-term conversion), using Conformal Prediction and data from ADNI. Despite the clinical meaning and significance of the proposed approach, the dataset used in this study did not allow a proper validation of the methodology, as more examples of long-term converting patients are required. Further work is needed to improve the methodology. In particular, we aim to test the approach with a larger data sample, either using different sources of data or using follow-up assessments to build learning examples, considering those as being a baseline assessment. In this case, we should guarantee that examples of the same patient are bundled together, either present in the train or in the test set. Besides that, we would like to test whether cases whose conversion is at the borderline of the time window (i.e., a patient that converts for instance at 2 years and more (or less) a few months) are introducing noise in the classification task. We will also try different cut-offs of the time windows, which may putatively be more significant clinically. Another future work concerns the use of methods to deal with multiclass problems, where the classifier should distinguish, in one step, which patients would convert in a short-term, long-term or remain stable. Different algorithms and data types should also be tested. We highlight the importance of introducing confidence levels associated with the predictions. Firstly, it is paramount for clinicians to know how much they can rely on the prediction made for a given patient, and the time that it takes for the conversion, in order to timely adjust the treatments. Secondly, even untrustworthy predictions might be useful as clinicians can prescribe more specific exams to deeply evaluate the neurodegeneration of these patients. 5. ACKNOWLEDGMENTS This work was partially supported by FCT under the Neuroclinomics2 project PTDC/EEI-SII/1937/2014, research grant (SFRH/ BD/95846/2013) to TP and FF (SFRH/BD/118872/2016), INESC-ID plurianual ref. UID/CEC/50021/2013, and LASIGE Research Unit ref. UID/CEC/00408/2013. Data collection and sharing for this project was funded by the Alzheimers Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH ). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimers Association; Alzheimers Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health ( The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimers Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. 6. REFERENCES [1] Barnes, D.E. et al A point-based tool to predict conversion from mild cognitive impairment to probable Alzheimer s disease. Alzheimer s & dementia : the journal of the Alzheimer s Association. 10, 6 (2014), DOI: [2] Cabral, C. et al Predicting conversion from MCI to AD with FDG-PET brain images at different prodromal stages. Computers in Biology and Medicine. 58, (2015), DOI: [3] Chawla, N. V et al SMOTE : Synthetic Minority Oversampling Technique. Journal of Artificial Intelligence Research. 16, (2002), [4] Demsar, J Statistical Comparisons of Classifiers over Multiple Data Sets. Journal of Machine Learning Research. 7, (2006), [5] Devetyarov, D. et al Conformal predictors in early diagnostics of ovarian and breast cancers. Progress in Artificial Intelligence. 1, 3 (2012), DOI: [6] Egerházi, A. et al Automated Neuropsychological Test Battery (CANTAB) in mild cognitive impairment and in Alzheimer s disease. Progress in neuro-psychopharmacology & biological psychiatry. 31, 3 (2007), DOI: [7] Eskildsen, S.F. et al Prediction of Alzheimer s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. NeuroImage. 65, (2013), DOI: n. [8] Grande, G. et al Reversible Mild Cognitive Impairment: The Role of Comorbidities at Baseline Evaluation. Journal of Alzheimer s Disease. 51, (2016),
7 [9] Hall, M.A Correlation-based Feature Selection for Machine Learning. University of Waikato. [10] Lambrou, A. et al Assessment of stroke risk based on morphological ultrasound image analysis with conformal prediction. Artificial Intelligence Applications and Innovations. (2010), DOI: 8. [11] Langa, K. and Levine, D The diagnosis and management of mild cognitive impairment: a clinical review. JAMA. 312, 23 (2014), DOI: [12] Lee, S.J. et al A clinical index to predict progression from mild cognitive impairment to dementia due to Alzheimer s disease. PloS one. 9, 12 (2014), e DOI: [13] Moradi, E. et al Machine learning framework for early MRI-based Alzheimer s conversion prediction in MCI subjects. NeuroImage. 104, (2014), DOI: [14] Norinder, U. et al Introducing Conformal Prediction in Predictive Modeling. A Transparent and Flexible Alternative to Applicability Domain Determination. Journal of Chemical Information and Modeling. 54, (2014), [15] Papadopoulos, H. et al Reliable diagnosis of acute abdominal pain with conformal prediction. Engineering Intelligent Systems. 17, 2 3 (2009), [16] Papadopoulos, H Reliable probabilistic prediction for medical decision support. IFIP Advances in Information and Communication Technology. 364 AICT, PART 2 (2011), DOI: [17] Pereira, T. et al Predicting the progression of mild cognitive impairment to dementia using neuropsychological data: a supervised learning approach using time windows. BMC Medical Informatics and Decision Making. 17:110, (2017). DOI: /s [18] Prince, M. et al World Alzheimer Report 2015: The Global Impact of Dementia - An analysis of prevalence, incidence, cost and trends. [19] Ribeiro, M.T. et al Why should I trust you? Explaining the predictions of any classifier. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) (2016), [20] Shafer, G. and Vovk, V A tutorial on conformal prediction. Journal of Machine Learning Research. 9, (2008), [21] Silva, D. et al Prediction of long-term (5 years) conversion to dementia using neuropsychological tests in a memory clinic setting. Journal of Alzheimer s disease : JAD. 34, 3 (2013), DOI: [22] Toccacheli, P. et al Conformal Predictors for Compound Activity Prediction. Conformal and Probabilistic Prediction with Applications [23] Vovk, V. et al Algorithmic Learning in a Random World. Springer. [24] Wolfsgruber, S. et al The CERAD neuropsychological assessment battery total score detects and predicts alzheimer disease dementia with high diagnostic accuracy. American Journal of Geriatric Psychiatry. 22, 10 (2014), DOI: Table 2. List of the 10 highest and lowest ranked subsets of features obtained with InfoGain method for Models 1 and 2. Highest ranked features Lowest ranked features Model 1 Model 2 Model 1 Model 2 ADAS-Cog - Delayed Word Recall ADAS-Cog - Total Score (ADAS 13) RAVLT - Trial B Total Age at retirement Trail Making Test - Part B NPI - Appetite and eating disorders: Item score MOCA - Letter Fluency RAVLT - Delayed Recall FAQ - Total Score MMSE - Total Score NPI - Total Score Category Fluency (Animals) - Total Correct ADAS-Cog - Word Recall ADAS-Cog - Orientation Score Education Level Trail Making Test - Part A ADAS-Cog - Total Score Years of symptoms prior to ADAS-Cog - Word Recall (ADAS 11) baseline FAQ - Total Score CDR - Home and Hobbies NPI - Depression/Dysphoria: Score Item score MOCA - Digit Span Forward NPI - Total Score CDR - Memory Score ADAS-Cog - Delayed Word Recall Gender MOCA - Delayed Recall RAVLT - Trial 6 Total NPI - Irritability/Lability: Item Diagnostic Summary - score Subjective memory complaint MMSE - Writing CDR - Judgment and Problem Solving Score RAVLT - Trial B Total NPI - Anxiety: Item score MOCA - Attention - Letters and Tapping MMSE - Construction - Copy Score Age at baseline Trail Making Test - Part B ADAS-Cog - Total Score (ADAS 11)
8 Table 4. Results obtained with our stepwise approach on the test set using the models trained with the optimized parameters and following the Forced Prediction (FP) approach. We note that smci stands for stable MCI, cmci: MCI, s.t. cmci: short-term conversion, l.t. cmci: long-term conversion and cfd: confidence threshold. smci Test set Model 1 Model 2 smci Nonclassifiable as cmci as smci cmci as cmci Misclassified cases (%) cmci Misclassified Nonclassifiable cases (%) cmci cfd Total cmci cmci Total cmci cmci cmci cmci ± ± ± ± ± ± ± ±2.6 14± ± ± ± ± ± ± ± ± ±1.9 18± ± ± ± ± ± ± ± ± ± ±11.7 Table 5. Results obtained with our stepwise approach on the test set using the models trained with the optimized parameters and following the Prediction Region (PR) approach. We note that smci stands for stable MCI, cmci: MCI, s.t. cmci: short-term conversion, l.t. cmci: long-term conversion and cfd: confidence threshold. Cfd. smci Test set Model 1 Model 2 smci Nonclassifiable cmci as cmci as cmci as cmci Misclassified cmci Misclassified smci cases (%) Total Total cmci cmci cmci cmci cmci cmci Nonclassifiable cases (%) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±15.1
AUTOMATING NEUROLOGICAL DISEASE DIAGNOSIS USING STRUCTURAL MR BRAIN SCAN FEATURES
AUTOMATING NEUROLOGICAL DISEASE DIAGNOSIS USING STRUCTURAL MR BRAIN SCAN FEATURES ALLAN RAVENTÓS AND MOOSA ZAIDI Stanford University I. INTRODUCTION Nine percent of those aged 65 or older and about one
More informationCHAPTER 5 NEUROPSYCHOLOGICAL PROFILE OF ALZHEIMER S DISEASE
CHAPTER 5 NEUROPSYCHOLOGICAL PROFILE OF ALZHEIMER S DISEASE 5.1 GENERAL BACKGROUND Neuropsychological assessment plays a crucial role in the assessment of cognitive decline in older age. In India, there
More informationData complexity measures for analyzing the effect of SMOTE over microarrays
ESANN 216 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Bruges (Belgium), 27-29 April 216, i6doc.com publ., ISBN 978-2878727-8. Data complexity
More informationSelection and Combination of Markers for Prediction
Selection and Combination of Markers for Prediction NACC Data and Methods Meeting September, 2010 Baojiang Chen, PhD Sarah Monsell, MS Xiao-Hua Andrew Zhou, PhD Overview 1. Research motivation 2. Describe
More informationPredicting Breast Cancer Survivability Rates
Predicting Breast Cancer Survivability Rates For data collected from Saudi Arabia Registries Ghofran Othoum 1 and Wadee Al-Halabi 2 1 Computer Science, Effat University, Jeddah, Saudi Arabia 2 Computer
More informationA Comparison of Collaborative Filtering Methods for Medication Reconciliation
A Comparison of Collaborative Filtering Methods for Medication Reconciliation Huanian Zheng, Rema Padman, Daniel B. Neill The H. John Heinz III College, Carnegie Mellon University, Pittsburgh, PA, 15213,
More informationInferior and medial temporal tau and cortical amyloid are associated with daily functional impairment in Alzheimer s disease
Halawa et al. Alzheimer's Research & Therapy (2019) 11:14 https://doi.org/10.1186/s13195-019-0471-6 RESEARCH Inferior and medial temporal tau and cortical amyloid are associated with daily functional impairment
More informationWhen Overlapping Unexpectedly Alters the Class Imbalance Effects
When Overlapping Unexpectedly Alters the Class Imbalance Effects V. García 1,2, R.A. Mollineda 2,J.S.Sánchez 2,R.Alejo 1,2, and J.M. Sotoca 2 1 Lab. Reconocimiento de Patrones, Instituto Tecnológico de
More informationAD Prevention Trials: An Industry Perspective
AD Prevention Trials: An Industry Perspective Robert A. Lasser, MD, MBA Global Development Lead, Product Development Group Medical Director, Neurodegenerative Disorders F. Hoffmann - La Roche Ltd Regulatory
More informationNEUROPSYCHOMETRIC TESTS
NEUROPSYCHOMETRIC TESTS CAMCOG It is the Cognitive section of Cambridge Examination for Mental Disorders of the Elderly (CAMDEX) The measure assesses orientation, language, memory, praxis, attention, abstract
More informationTHE ROLE OF ACTIVITIES OF DAILY LIVING IN THE MCI SYNDROME
PERNECZKY 15/06/06 14:35 Page 1 THE ROLE OF ACTIVITIES OF DAILY LIVING IN THE MCI SYNDROME R. PERNECZKY, A. KURZ Department of Psychiatry and Psychotherapy, Technical University of Munich, Germany. Correspondence
More informationDevelopment of Soft-Computing techniques capable of diagnosing Alzheimer s Disease in its pre-clinical stage combining MRI and FDG-PET images.
Development of Soft-Computing techniques capable of diagnosing Alzheimer s Disease in its pre-clinical stage combining MRI and FDG-PET images. Olga Valenzuela, Francisco Ortuño, Belen San-Roman, Victor
More informationDiagnosis of Breast Cancer Using Ensemble of Data Mining Classification Methods
International Journal of Bioinformatics and Biomedical Engineering Vol. 1, No. 3, 2015, pp. 318-322 http://www.aiscience.org/journal/ijbbe ISSN: 2381-7399 (Print); ISSN: 2381-7402 (Online) Diagnosis of
More informationHeterogeneous Data Mining for Brain Disorder Identification. Bokai Cao 04/07/2015
Heterogeneous Data Mining for Brain Disorder Identification Bokai Cao 04/07/2015 Outline Introduction Tensor Imaging Analysis Brain Network Analysis Davidson et al. Network discovery via constrained tensor
More informationDetection of Mild Cognitive Impairment using Image Differences and Clinical Features
Detection of Mild Cognitive Impairment using Image Differences and Clinical Features L I N L I S C H O O L O F C O M P U T I N G C L E M S O N U N I V E R S I T Y Copyright notice Many of the images in
More informationWHI Memory Study (WHIMS) Investigator Data Release Data Preparation Guide December 2012
WHI Memory Study (WHIMS) Investigator Data Release Data Preparation Guide December 2012 1. Introduction Changes in the current update (December 2012): New data sets Post Trial - Form A, Phase 2: Administration
More informationAn Improved Algorithm To Predict Recurrence Of Breast Cancer
An Improved Algorithm To Predict Recurrence Of Breast Cancer Umang Agrawal 1, Ass. Prof. Ishan K Rajani 2 1 M.E Computer Engineer, Silver Oak College of Engineering & Technology, Gujarat, India. 2 Assistant
More informationEvaluating Classifiers for Disease Gene Discovery
Evaluating Classifiers for Disease Gene Discovery Kino Coursey Lon Turnbull khc0021@unt.edu lt0013@unt.edu Abstract Identification of genes involved in human hereditary disease is an important bioinfomatics
More informationPredicting Kidney Cancer Survival from Genomic Data
Predicting Kidney Cancer Survival from Genomic Data Christopher Sauer, Rishi Bedi, Duc Nguyen, Benedikt Bünz Abstract Cancers are on par with heart disease as the leading cause for mortality in the United
More informationarxiv: v1 [cs.lg] 4 Feb 2019
Machine Learning for Seizure Type Classification: Setting the benchmark Subhrajit Roy [000 0002 6072 5500], Umar Asif [0000 0001 5209 7084], Jianbin Tang [0000 0001 5440 0796], and Stefan Harrer [0000
More informationImproved Intelligent Classification Technique Based On Support Vector Machines
Improved Intelligent Classification Technique Based On Support Vector Machines V.Vani Asst.Professor,Department of Computer Science,JJ College of Arts and Science,Pudukkottai. Abstract:An abnormal growth
More informationBioArctic announces detailed results of the BAN2401 Phase 2b study in early Alzheimer s disease presented at AAIC 2018
Press release BioArctic announces detailed results of the BAN2401 Phase 2b study in early Alzheimer s disease presented at AAIC 2018 Stockholm, Sweden, July 25, 2018 BioArctic AB (publ) (Nasdaq Stockholm:
More informationBinary Classification of Cognitive Disorders: Investigation on the Effects of Protein Sequence Properties in Alzheimer s and Parkinson s Disease
Binary Classification of Cognitive Disorders: Investigation on the Effects of Protein Sequence Properties in Alzheimer s and Parkinson s Disease Shomona Gracia Jacob, Member, IAENG, Tejeswinee K Abstract
More informationAssigning B cell Maturity in Pediatric Leukemia Gabi Fragiadakis 1, Jamie Irvine 2 1 Microbiology and Immunology, 2 Computer Science
Assigning B cell Maturity in Pediatric Leukemia Gabi Fragiadakis 1, Jamie Irvine 2 1 Microbiology and Immunology, 2 Computer Science Abstract One method for analyzing pediatric B cell leukemia is to categorize
More informationComparing Multifunctionality and Association Information when Classifying Oncogenes and Tumor Suppressor Genes
000 001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022 023 024 025 026 027 028 029 030 031 032 033 034 035 036 037 038 039 040 041 042 043 044 045 046 047 048 049 050
More informationInvestigating the performance of a CAD x scheme for mammography in specific BIRADS categories
Investigating the performance of a CAD x scheme for mammography in specific BIRADS categories Andreadis I., Nikita K. Department of Electrical and Computer Engineering National Technical University of
More informationThe Australian Imaging Biomarkers and Lifestyle Flagship Study of Ageing
The Australian Imaging Biomarkers and Lifestyle Flagship Study of Ageing. (AUSTRALIAN ADNI) July 2013 UPDATE Imaging Christopher Rowe MD Neuroimaging stream leader June 2013 The Australian Imaging Biomarkers
More informationAn integrated natural disease progression model of nine cognitive and biomarker endpoints in patients with Alzheimer s Disease
An integrated natural disease progression model of nine cognitive and biomarker endpoints in patients with Alzheimer s Disease Angelica Quartino 1* Dan Polhamus 2*, James Rogers 2, Jin Jin 1 212, Genentech
More informationMild Cognitive Impairment (MCI)
October 19, 2018 Mild Cognitive Impairment (MCI) Yonas E. Geda, MD, MSc Professor of Neurology and Psychiatry Consultant, Departments of Psychiatry & Psychology, and Neurology Mayo Clinic College of Medicine
More informationUDS version 3 Summary of major changes to UDS form packets
UDS version 3 Summary of major changes to UDS form packets from version 2 to VERSION 3 february 18 final Form A1: Subject demographics Updated question on principal referral source to add additional options
More information2013 AAIC WW-ADNI Meeting, Boston K-ADNI Update
2013 AAIC WW-ADNI Meeting, Boston K-ADNI Update July 12th, 2013 Seong Yoon Kim, Seol Hee Han, Duk L. Na Update for K-ADNI, 2013 General Year1 Tasks Infrastructure setup for K-ADNI is under way. Database
More informationMultilayer clustering: Biomarker driven segmentation of Alzheimer s disease patient population
Multilayer clustering: Biomarker driven segmentation of Alzheimer s disease patient population Dragan Gamberger Rudjer Bošković Institute, Zagreb, Croatia Bernard Ženko Jožef Stefan Institute, Ljubljana,
More informationPositive and Unlabeled Relational Classification through Label Frequency Estimation
Positive and Unlabeled Relational Classification through Label Frequency Estimation Jessa Bekker and Jesse Davis Computer Science Department, KU Leuven, Belgium firstname.lastname@cs.kuleuven.be Abstract.
More informationOverview. Case #1 4/20/2012. Neuropsychological assessment of older adults: what, when and why?
Neuropsychological assessment of older adults: what, when and why? Benjamin Mast, Ph.D. Associate Professor & Vice Chair, Psychological & Brain Sciences Associate Clinical Professor, Family & Geriatric
More informationPredictive performance and discrimination in unbalanced classification
MASTER Predictive performance and discrimination in unbalanced classification van der Zon, S.B. Award date: 2016 Link to publication Disclaimer This document contains a student thesis (bachelor's or master's),
More informationCLASSIFICATION OF BRAIN TUMOUR IN MRI USING PROBABILISTIC NEURAL NETWORK
CLASSIFICATION OF BRAIN TUMOUR IN MRI USING PROBABILISTIC NEURAL NETWORK PRIMI JOSEPH (PG Scholar) Dr.Pauls Engineering College Er.D.Jagadiswary Dr.Pauls Engineering College Abstract: Brain tumor is an
More informationPerformance Evaluation of Machine Learning Algorithms in the Classification of Parkinson Disease Using Voice Attributes
Performance Evaluation of Machine Learning Algorithms in the Classification of Parkinson Disease Using Voice Attributes J. Sujatha Research Scholar, Vels University, Assistant Professor, Post Graduate
More informationPositive and Unlabeled Relational Classification through Label Frequency Estimation
Positive and Unlabeled Relational Classification through Label Frequency Estimation Jessa Bekker and Jesse Davis Computer Science Department, KU Leuven, Belgium firstname.lastname@cs.kuleuven.be Abstract.
More informationInter-session reproducibility measures for high-throughput data sources
Inter-session reproducibility measures for high-throughput data sources Milos Hauskrecht, PhD, Richard Pelikan, MSc Computer Science Department, Intelligent Systems Program, Department of Biomedical Informatics,
More informationUNIVERSITY of PENNSYLVANIA CIS 520: Machine Learning Final, Fall 2014
UNIVERSITY of PENNSYLVANIA CIS 520: Machine Learning Final, Fall 2014 Exam policy: This exam allows two one-page, two-sided cheat sheets (i.e. 4 sides); No other materials. Time: 2 hours. Be sure to write
More informationLogistic Regression and Bayesian Approaches in Modeling Acceptance of Male Circumcision in Pune, India
20th International Congress on Modelling and Simulation, Adelaide, Australia, 1 6 December 2013 www.mssanz.org.au/modsim2013 Logistic Regression and Bayesian Approaches in Modeling Acceptance of Male Circumcision
More informationA NOVEL VARIABLE SELECTION METHOD BASED ON FREQUENT PATTERN TREE FOR REAL-TIME TRAFFIC ACCIDENT RISK PREDICTION
OPT-i An International Conference on Engineering and Applied Sciences Optimization M. Papadrakakis, M.G. Karlaftis, N.D. Lagaros (eds.) Kos Island, Greece, 4-6 June 2014 A NOVEL VARIABLE SELECTION METHOD
More informationPredicting Sleep Using Consumer Wearable Sensing Devices
Predicting Sleep Using Consumer Wearable Sensing Devices Miguel A. Garcia Department of Computer Science Stanford University Palo Alto, California miguel16@stanford.edu 1 Introduction In contrast to the
More informationIdentifying Parkinson s Patients: A Functional Gradient Boosting Approach
Identifying Parkinson s Patients: A Functional Gradient Boosting Approach Devendra Singh Dhami 1, Ameet Soni 2, David Page 3, and Sriraam Natarajan 1 1 Indiana University Bloomington 2 Swarthmore College
More informationAn empirical evaluation of text classification and feature selection methods
ORIGINAL RESEARCH An empirical evaluation of text classification and feature selection methods Muazzam Ahmed Siddiqui Department of Information Systems, Faculty of Computing and Information Technology,
More informationBioArctic announces positive topline results of BAN2401 Phase 2b at 18 months in early Alzheimer s Disease
Press release BioArctic announces positive topline results of BAN2401 Phase 2b at 18 months in early Alzheimer s Disease The full 18 month analysis of the 856 patient BAN2401 Phase 2b clinical study in
More informationApplication of Artificial Neural Networks in Classification of Autism Diagnosis Based on Gene Expression Signatures
Application of Artificial Neural Networks in Classification of Autism Diagnosis Based on Gene Expression Signatures 1 2 3 4 5 Kathleen T Quach Department of Neuroscience University of California, San Diego
More informationInternational Journal of Pharma and Bio Sciences A NOVEL SUBSET SELECTION FOR CLASSIFICATION OF DIABETES DATASET BY ITERATIVE METHODS ABSTRACT
Research Article Bioinformatics International Journal of Pharma and Bio Sciences ISSN 0975-6299 A NOVEL SUBSET SELECTION FOR CLASSIFICATION OF DIABETES DATASET BY ITERATIVE METHODS D.UDHAYAKUMARAPANDIAN
More informationContributions to Brain MRI Processing and Analysis
Contributions to Brain MRI Processing and Analysis Dissertation presented to the Department of Computer Science and Artificial Intelligence By María Teresa García Sebastián PhD Advisor: Prof. Manuel Graña
More informationGene Selection for Tumor Classification Using Microarray Gene Expression Data
Gene Selection for Tumor Classification Using Microarray Gene Expression Data K. Yendrapalli, R. Basnet, S. Mukkamala, A. H. Sung Department of Computer Science New Mexico Institute of Mining and Technology
More informationPerformance Analysis of Different Classification Methods in Data Mining for Diabetes Dataset Using WEKA Tool
Performance Analysis of Different Classification Methods in Data Mining for Diabetes Dataset Using WEKA Tool Sujata Joshi Assistant Professor, Dept. of CSE Nitte Meenakshi Institute of Technology Bangalore,
More informationData Mining Techniques to Predict Survival of Metastatic Breast Cancer Patients
Data Mining Techniques to Predict Survival of Metastatic Breast Cancer Patients Abstract Prognosis for stage IV (metastatic) breast cancer is difficult for clinicians to predict. This study examines the
More informationMRI Image Processing Operations for Brain Tumor Detection
MRI Image Processing Operations for Brain Tumor Detection Prof. M.M. Bulhe 1, Shubhashini Pathak 2, Karan Parekh 3, Abhishek Jha 4 1Assistant Professor, Dept. of Electronics and Telecommunications Engineering,
More informationLearning with Rare Cases and Small Disjuncts
Appears in Proceedings of the 12 th International Conference on Machine Learning, Morgan Kaufmann, 1995, 558-565. Learning with Rare Cases and Small Disjuncts Gary M. Weiss Rutgers University/AT&T Bell
More informationMachine Learning Algorithms for Neuroimaging-based Clinical Trials in Preclinical Alzheimer s Disease
Machine Learning Algorithms for Neuroimaging-based Clinical Trials in Preclinical Alzheimer s Disease Vamsi K. Ithapu Wisconsin Alzheimer s Disease Research Center University of Wisconsin-Madison April
More informationand errs as expected. The disadvantage of this approach is that it is time consuming, due to the fact that it is necessary to evaluate all algorithms,
Data transformation and model selection by experimentation and meta-learning Pavel B. Brazdil LIACC, FEP - University of Porto Rua Campo Alegre, 823 4150 Porto, Portugal Email: pbrazdil@ncc.up.pt Research
More informationCognitive Screening in Risk Assessment. Geoffrey Tremont, Ph.D. Rhode Island Hospital & Alpert Medical School of Brown University.
Cognitive Screening in Risk Assessment Geoffrey Tremont, Ph.D. Rhode Island Hospital & Alpert Medical School of Brown University Outline of Talk Definition of Dementia and MCI Incidence and Prevalence
More informationA DATA MINING APPROACH FOR PRECISE DIAGNOSIS OF DENGUE FEVER
A DATA MINING APPROACH FOR PRECISE DIAGNOSIS OF DENGUE FEVER M.Bhavani 1 and S.Vinod kumar 2 International Journal of Latest Trends in Engineering and Technology Vol.(7)Issue(4), pp.352-359 DOI: http://dx.doi.org/10.21172/1.74.048
More informationIdentifying Signs of Depression on Twitter Eugene Tang, Class of 2016 Dobin Prize Submission
Identifying Signs of Depression on Twitter Eugene Tang, Class of 2016 Dobin Prize Submission INTRODUCTION Depression is estimated to affect 350 million people worldwide (WHO, 2015). Characterized by feelings
More informationUsing AUC and Accuracy in Evaluating Learning Algorithms
1 Using AUC and Accuracy in Evaluating Learning Algorithms Jin Huang Charles X. Ling Department of Computer Science The University of Western Ontario London, Ontario, Canada N6A 5B7 fjhuang, clingg@csd.uwo.ca
More informationComparative Analysis of Machine Learning Algorithms for Chronic Kidney Disease Detection using Weka
I J C T A, 10(8), 2017, pp. 59-67 International Science Press ISSN: 0974-5572 Comparative Analysis of Machine Learning Algorithms for Chronic Kidney Disease Detection using Weka Milandeep Arora* and Ajay
More informationA Survey on Brain Tumor Detection Technique
(International Journal of Computer Science & Management Studies) Vol. 15, Issue 06 A Survey on Brain Tumor Detection Technique Manju Kadian 1 and Tamanna 2 1 M.Tech. Scholar, CSE Department, SPGOI, Rohtak
More informationA NEW DIAGNOSIS SYSTEM BASED ON FUZZY REASONING TO DETECT MEAN AND/OR VARIANCE SHIFTS IN A PROCESS. Received August 2010; revised February 2011
International Journal of Innovative Computing, Information and Control ICIC International c 2011 ISSN 1349-4198 Volume 7, Number 12, December 2011 pp. 6935 6948 A NEW DIAGNOSIS SYSTEM BASED ON FUZZY REASONING
More informationAlzheimer s Disease Risk Assessment Using Large-Scale Machine Learning Methods
Alzheimer s Disease Risk Assessment Using Large-Scale Machine Learning Methods Ramon Casanova*, Fang-Chi Hsu, Kaycee M. Sink, Stephen R. Rapp, Jeff D. Williamson, Susan M. Resnick, Mark A. Espeland, for
More informationClassification and Statistical Analysis of Auditory FMRI Data Using Linear Discriminative Analysis and Quadratic Discriminative Analysis
International Journal of Innovative Research in Computer Science & Technology (IJIRCST) ISSN: 2347-5552, Volume-2, Issue-6, November-2014 Classification and Statistical Analysis of Auditory FMRI Data Using
More informationCustom Intelligence. Alzheimer s Disease Landscape Summary
Custom Intelligence Alzheimer s Disease Landscape Summary Prepared August 21 Alzheimer s Disease Landscape: Executive Summary Current standard of care in Alzheimer s Disease (AD) consists of symptomatic
More informationThe current state of healthcare for Normal Aging, Mild Cognitive Impairment, & Alzheimer s Disease
The current state of healthcare for Normal Aging, g, Mild Cognitive Impairment, & Alzheimer s Disease William Rodman Shankle, MS MD FACP Director, Alzheimer s Program, Hoag Neurosciences Institute Neurologist,
More informationMinority Report: ML Fairness in Criminality Prediction
Minority Report: ML Fairness in Criminality Prediction Dominick Lim djlim@stanford.edu Torin Rudeen torinmr@stanford.edu 1. Introduction 1.1. Motivation Machine learning is used more and more to make decisions
More informationDo Data Mining Methods Support the Three-Group Diagnostic Model of Primary Progressive Aphasia?
Do Data Mining Methods Support the ThreeGroup Diagnostic Model of Primary Progressive Aphasia? Telma Pereira * tpereira@kdbio.inescid.pt Carolina Maruta carolmaruta@gmail.com Alexandre de Mendonça mendonca@medicina.ulisboa.pt
More informationIdentifying Thyroid Carcinoma Subtypes and Outcomes through Gene Expression Data Kun-Hsing Yu, Wei Wang, Chung-Yu Wang
Identifying Thyroid Carcinoma Subtypes and Outcomes through Gene Expression Data Kun-Hsing Yu, Wei Wang, Chung-Yu Wang Abstract: Unlike most cancers, thyroid cancer has an everincreasing incidence rate
More informationBiomarker adaptive designs in clinical trials
Review Article Biomarker adaptive designs in clinical trials James J. Chen 1, Tzu-Pin Lu 1,2, Dung-Tsa Chen 3, Sue-Jane Wang 4 1 Division of Bioinformatics and Biostatistics, National Center for Toxicological
More informationPredicting Breast Cancer Survival Using Treatment and Patient Factors
Predicting Breast Cancer Survival Using Treatment and Patient Factors William Chen wchen808@stanford.edu Henry Wang hwang9@stanford.edu 1. Introduction Breast cancer is the leading type of cancer in women
More informationEarly Diagnosis of Alzheimer s Disease and MCI via Imaging and Pattern Analysis Methods. Christos Davatzikos, Ph.D.
Early Diagnosis of Alzheimer s Disease and MCI via Imaging and Pattern Analysis Methods Christos Davatzikos, Ph.D. Director, Section of Biomedical Image Analysis Professor of Radiology http://www.rad.upenn.edu/sbia
More informationMild Cognitive Impairment
Mild Cognitive Impairment Victor W. Henderson, MD, MS Departments of Health Research & Policy (Epidemiology) and of Neurology & Neurological Sciences Stanford University Director, Stanford Alzheimer s
More informationFOR IMMEDIATE RELEASE July 26, 2018 Eisai Co., Ltd. Biogen Inc.
FOR IMMEDIATE RELEASE July 26, 2018 Eisai Co., Ltd. Biogen Inc. EISAI AND BIOGEN ANNOUNCE DETAILED RESULTS OF PHASE II CLINICAL STUDY OF BAN2401 IN EARLY ALZHEIMER S DISEASE AT ALZHEIMER S ASSOCIATION
More informationJournal of Advanced Scientific Research ROUGH SET APPROACH FOR FEATURE SELECTION AND GENERATION OF CLASSIFICATION RULES OF HYPOTHYROID DATA
Kavitha et al., J Adv Sci Res, 2016, 7(2): 15-19 15 Journal of Advanced Scientific Research Available online through http://www.sciensage.info/jasr ISSN 0976-9595 Research Article ROUGH SET APPROACH FOR
More informationABSTRACT I. INTRODUCTION. Mohd Thousif Ahemad TSKC Faculty Nagarjuna Govt. College(A) Nalgonda, Telangana, India
International Journal of Scientific Research in Computer Science, Engineering and Information Technology 2018 IJSRCSEIT Volume 3 Issue 1 ISSN : 2456-3307 Data Mining Techniques to Predict Cancer Diseases
More informationEXPEDITION3: A Phase 3 Trial of Solanezumab in Mild Dementia due to Alzheimer s Disease
EXPEDITION3: A Phase 3 Trial of in Mild Dementia due to Alzheimer s Disease Lawrence S. Honig, MD, PhD On behalf of the EXPEDITION3 Study Team Disclosure Statement I will discuss investigational use only.
More informationDetection of Cognitive States from fmri data using Machine Learning Techniques
Detection of Cognitive States from fmri data using Machine Learning Techniques Vishwajeet Singh, K.P. Miyapuram, Raju S. Bapi* University of Hyderabad Computational Intelligence Lab, Department of Computer
More informationMild Cognitive Impairment Symposium January 19 and 20, 2013
Highlights of Biomarker and Clinical Outcomes in Recent AD Treatment Trials Stephen Salloway, MD, MS Professor of Neurology and Psychiatry Alpert Medical School, Brown University Mild Cognitive Impairment
More informationAnalysis of Classification Algorithms towards Breast Tissue Data Set
Analysis of Classification Algorithms towards Breast Tissue Data Set I. Ravi Assistant Professor, Department of Computer Science, K.R. College of Arts and Science, Kovilpatti, Tamilnadu, India Abstract
More informationError Detection based on neural signals
Error Detection based on neural signals Nir Even- Chen and Igor Berman, Electrical Engineering, Stanford Introduction Brain computer interface (BCI) is a direct communication pathway between the brain
More informationHow can the new diagnostic criteria improve patient selection for DM therapy trials
How can the new diagnostic criteria improve patient selection for DM therapy trials Amsterdam, August 2015 Bruno Dubois Head of the Dementia Research Center (IMMA) Director of INSERM Research Unit (ICM)
More informationECG Beat Recognition using Principal Components Analysis and Artificial Neural Network
International Journal of Electronics Engineering, 3 (1), 2011, pp. 55 58 ECG Beat Recognition using Principal Components Analysis and Artificial Neural Network Amitabh Sharma 1, and Tanushree Sharma 2
More informationMining Low-Support Discriminative Patterns from Dense and High-Dimensional Data. Technical Report
Mining Low-Support Discriminative Patterns from Dense and High-Dimensional Data Technical Report Department of Computer Science and Engineering University of Minnesota 4-192 EECS Building 200 Union Street
More informationLatest Methods to Early Detection for Alzheimer's: Cognitive Assessments and Diagnostic Tools in Practice
Transcript Details This is a transcript of an educational program accessible on the ReachMD network. Details about the program and additional media formats for the program are accessible by visiting: https://reachmd.com/programs/alzheimers-disease-towards-earlier-detection/latest-methods-earlydetection-alzheimers-cognitive-assessments-and-diagnostic-tools-practice/8321/
More informationPrediction of Diabetes Using Bayesian Network
Prediction of Diabetes Using Bayesian Network Mukesh kumari 1, Dr. Rajan Vohra 2,Anshul arora 3 1,3 Student of M.Tech (C.E) 2 Head of Department Department of computer science & engineering P.D.M College
More informationData Mining in Bioinformatics Day 4: Text Mining
Data Mining in Bioinformatics Day 4: Text Mining Karsten Borgwardt February 25 to March 10 Bioinformatics Group MPIs Tübingen Karsten Borgwardt: Data Mining in Bioinformatics, Page 1 What is text mining?
More informationIdentification of Neuroimaging Biomarkers
Identification of Neuroimaging Biomarkers Dan Goodwin, Tom Bleymaier, Shipra Bhal Advisor: Dr. Amit Etkin M.D./PhD, Stanford Psychiatry Department Abstract We present a supervised learning approach to
More informationWeighted Naive Bayes Classifier: A Predictive Model for Breast Cancer Detection
Weighted Naive Bayes Classifier: A Predictive Model for Breast Cancer Detection Shweta Kharya Bhilai Institute of Technology, Durg C.G. India ABSTRACT In this paper investigation of the performance criterion
More informationChapter 1. Introduction
Chapter 1 Introduction 1.1 Motivation and Goals The increasing availability and decreasing cost of high-throughput (HT) technologies coupled with the availability of computational tools and data form a
More informationTITLE: A Data-Driven Approach to Patient Risk Stratification for Acute Respiratory Distress Syndrome (ARDS)
TITLE: A Data-Driven Approach to Patient Risk Stratification for Acute Respiratory Distress Syndrome (ARDS) AUTHORS: Tejas Prahlad INTRODUCTION Acute Respiratory Distress Syndrome (ARDS) is a condition
More informationIan McKeith MD, F Med Sci, Professor of Old Age Psychiatry, Newcastle University
Ian McKeith MD, F Med Sci, Professor of Old Age Psychiatry, Newcastle University Design of trials in DLB and PDD What has been learnt from previous trials in these indications and other dementias? Overview
More informationResearch on Classification of Diseases of Clinical Imbalanced Data in Traditional Chinese Medicine
Int'l Conf. Bioinformatics and Computational Biology BIOCOMP'17 57 Research on Classification of Diseases of Clinical Imbalanced Data in Traditional Chinese Medicine Zhu-Qiang Pan School of Computer Science,
More informationEXTRACT THE BREAST CANCER IN MAMMOGRAM IMAGES
International Journal of Civil Engineering and Technology (IJCIET) Volume 10, Issue 02, February 2019, pp. 96-105, Article ID: IJCIET_10_02_012 Available online at http://www.iaeme.com/ijciet/issues.asp?jtype=ijciet&vtype=10&itype=02
More information, Denis Mariano-Goulart 3, Pierre Payoux 1,2 and the Alzheimer s Disease Neuroimaging. Fayçal Ben Bouallègue 1,2,3* Initiative (ADNI)
Ben Bouallègue et al. Alzheimer's Research & Therapy (2017) 9:32 DOI 10.1186/s13195-017-0260-z RESEARCH Comparison of CSF markers and semi-quantitative amyloid PET in Alzheimer s disease diagnosis and
More informationWHI Memory Study (WHIMS) Investigator Data Release Data Preparation Guide April 2014
WHI Memory Study (WHIMS) Investigator Data Release Data Preparation Guide April 2014 1. Introduction This release consists of a single data set from the WHIMS Epidemiology of Cognitive Health Outcomes
More informationMild Behavioral Impairment (MBI): Symptoms, Prodrome, or False Alarm?
Mild Behavioral Impairment (MBI): Symptoms, Prodrome, or False Alarm? Constantine G. Lyketsos, MD, MHS Chair of Psychiatry, Johns Hopkins Bayview Elizabeth Plank Althouse Professor, Johns Hopkins University
More informationEnhanced Detection of Lung Cancer using Hybrid Method of Image Segmentation
Enhanced Detection of Lung Cancer using Hybrid Method of Image Segmentation L Uma Maheshwari Department of ECE, Stanley College of Engineering and Technology for Women, Hyderabad - 500001, India. Udayini
More informationDiscovering Meaningful Cut-points to Predict High HbA1c Variation
Proceedings of the 7th INFORMS Workshop on Data Mining and Health Informatics (DM-HI 202) H. Yang, D. Zeng, O. E. Kundakcioglu, eds. Discovering Meaningful Cut-points to Predict High HbAc Variation Si-Chi
More information